This morning I watched a short film by Gabriela Tropia where she collaborates on a series of short films and speaks to an AI recreation of the filmmaker Maya Deren (who died in 1961). It’s an exciting and inspirational piece of art that plays with boundaries of what AI is.
This week our AI health coach team has been replaying some of the research. The health coach team is testing the proposition – how will people react to AI giving them support to lose weight? So far the reaction to an AI health coach is positive. As a very early iteration, the user is shown a set of four coaches they can choose from. The coaches are represented by friendly but corporate illustrations with simple explanations of their coaching styles. The user talks with them using a text message style interface.
Both projects explore the vernacular of artificial humans. It’s interesting to think about the representation of AI as a new tool for designers, with its own language. For the AI health coach we need to understand if we can use the grammar of human interaction – empathy, consistency, encouragement – to improve outcomes.
Gabby’s film makes me question the assumptions we’ve made when designing the coach. Is thinking of AI as a person something to lean in to? Would a richer, more human‑feeling AI health coach improve health outcomes? Is there a form of artificial humanisation that builds trust? Is there an optimal level of authentically human? Too human and it might be unsafe. Too safe and it might not be human enough.
It feels likely that having more realistic visualisations offers greater authenticity and therefore is more compelling to the user. A more lifelike representation has more ways to communicate and is theoretically more familiar. But that has new challenges. Can we design clinically safe facial expressions? What if the AI is saying the things we intended but sounds really sarcastic?
In Gabby’s film Maya Deren goes off‑script and feels unpredictable, treading the line between being a wilful human and an AI hallucination. Gabby is making the viewer question what they’re seeing and inviting us to think about the construction on screen. Trust is a big component of what we can offer. The last thing we want is to erode people’s trust in Better Health or the NHS.
A simple illustration is an honest signifier of what the technology can reliably do. Simple illustrations give the user a shorthand for what to expect from the coach we’re building. A photorealistic coach would promise a richness we cannot offer.
Over time the vernacular of AI is going to develop. Text‑based chat and flat illustrations may well be tropes we lose as our experience of designing with AI grows and the technology improves.